Analysis

Critical transitions in the Amazon forest

system

https://doi.org/10.1038/s41586-023-06970-0 Bernardo M. Flores1 ✉, Encarni Montoya2, Boris Sakschewski3, Nathália Nascimento4,

Arie Staal5, Richard A. Betts6,7, Carolina Levis1, David M. Lapola8, Adriane Esquível-Muelbert9,10,
Received: 29 August 2022
Catarina Jakovac11, Carlos A. Nobre4, Rafael S. Oliveira12, Laura S. Borma13, Da Nian3,
Accepted: 13 December 2023 Niklas Boers3,14, Susanna B. Hecht15, Hans ter Steege16,17, Julia Arieira18, Isabella L. Lucas19,
Erika Berenguer20, José A. Marengo21,22,23, Luciana V. Gatti13, Caio R. C. Mattos24 &
Published online: 14 February 2024
Marina Hirota1,12,25 ✉
Open access

Check for updates

The possibility that the Amazon forest system could soon reach a tipping point,
inducing large-scale collapse, has raised global concern1–3. For 65 million years,
Amazonian forests remained relatively resilient to climatic variability. Now, the
region is increasingly exposed to unprecedented stress from warming temperatures,
extreme droughts, deforestation and fires, even in central and remote parts of the
system1. Long existing feedbacks between the forest and environmental conditions
are being replaced by novel feedbacks that modify ecosystem resilience, increasing
the risk of critical transition. Here we analyse existing evidence for five major drivers
of water stress on Amazonian forests, as well as potential critical thresholds of those
drivers that, if crossed, could trigger local, regional or even biome-wide forest
collapse. By combining spatial information on various disturbances, we estimate
that by 2050, 10% to 47% of Amazonian forests will be exposed to compounding
disturbances that may trigger unexpected ecosystem transitions and potentially
exacerbate regional climate change. Using examples of disturbed forests across the
Amazon, we identify the three most plausible ecosystem trajectories, involving
different feedbacks and environmental conditions. We discuss how the inherent
complexity of the Amazon adds uncertainty about future dynamics, but also reveals
opportunities for action. Keeping the Amazon forest resilient in the Anthropocene
will depend on a combination of local efforts to end deforestation and degradation
and to expand restoration, with global efforts to stop greenhouse gas emissions.

The Amazon forest is a complex system of interconnected species,
ecosystems and human cultures that contributes to the well-being of
people globally1. The Amazon forest holds more than 10% of Earth’s
terrestrial biodiversity, stores an amount of carbon equivalent to 15–20
years of global CO2 emissions (150–200 Pg C), and has a net cooling
effect (from evapotranspiration) that helps to stabilize the Earth’s
climate1–3. The forest contributes up to 50% of rainfall in the region
and is crucial for moisture supply across South America4, allowing
other biomes and economic activities to thrive in regions that would
otherwise be more arid, such as the Pantanal wetlands and the La Plata
river basin1. Large parts of the Amazon forest, however, are projected to
experience mass mortality events due to climatic and land use-related
disturbances in the coming decades5,6, potentially accelerating
climate change through carbon emissions and feedbacks with the
climate system2,3. These impacts would also involve irreversible loss
of biodiversity, socioeconomic and cultural values1,7–9. The Amazon
is home to more than 40 million people, including 2.2 million Indig-
enous peoples of more than 300 ethnicities, as well as afrodescend-
ent and local traditional communities1. Indigenous peoples and local
communities (IPLCs) would be harmed by forest loss in terms of their
livelihoods, lifeways and knowledge systems that inspire societies
globally1,7,9.

1
Graduate Program in Ecology, Federal University of Santa Catarina, Florianopolis, Brazil. 2Geosciences Barcelona, Spanish National Research Council, Barcelona, Spain. 3Potsdam Institute for
Climate Impact Research, Member of the Leibniz Association, Potsdam, Germany. 4Institute of Advanced Studies, University of São Paulo, São Paulo, Brazil. 5Copernicus Institute of Sustainable
Development, Utrecht University, Utrecht, The Netherlands. 6Met Office Hadley Centre, Exeter, UK. 7Global Systems Institute, University of Exeter, Exeter, UK. 8Center for Meteorological and
Climatic Research Applied to Agriculture, University of Campinas, Campinas, Brazil. 9School of Geography, Earth and Environmental Sciences, University of Birmingham, Birmingham, UK.
10
Birmingham Institute of Forest Research, University of Birmingham, Birmingham, UK. 11Department of Plant Sciences, Federal University of Santa Catarina, Florianopolis, Brazil. 12Department
of Plant Biology, University of Campinas, Campinas, Brazil. 13Division of Impacts, Adaptation and Vulnerabilities (DIIAV), National Institute for Space Research, São José dos Campos, Brazil.
14
Earth System Modelling, School of Engineering and Design, Technical University of Munich, Munich, Germany. 15Luskin School for Public Affairs and Institute of the Environment, University of
California, Los Angeles, CA, USA. 16Naturalis Biodiversity Center, Leiden, The Netherlands. 17Quantitative Biodiversity Dynamics, Utrecht University, Utrecht, The Netherlands. 18Science Panel for
the Amazon (SPA), São José dos Campos, Brazil. 19Sustainable Development Solutions Network, New York, NY, USA. 20Environmental Change Institute, University of Oxford, Oxford, UK.
21
Centro Nacional de Monitoramento e Alerta de Desastres Naturais, São José dos Campos, Brazil. 22Graduate Program in Natural Disasters, UNESP/CEMADEN, São José dos Campos, Brazil.
Graduate School of International Studies, Korea University, Seoul, Korea. 24Program in Atmospheric and Oceanic Sciences, Princeton University, Princeton, NJ, USA. 25Group IpES, Department
23

of Physics, Federal University of Santa Catarina, Florianopolis, Brazil. ✉e-mail: mflores.bernardo@gmail.com; marinahirota@gmail.com

Nature | Vol 626 | 15 February 2024 | 555

Analysis
Understanding the risk of such catastrophic behaviour requires
addressing complex factors that shape ecosystem resilience10. A major
question is whether a large-scale collapse of the Amazon forest system
could actually happen within the twenty-first century, and if this would
be associated with a particular tipping point. Here we synthesize evi-
dence from paleorecords, observational data and modelling studies of
critical drivers of stress on the system. We assess potential thresholds
of those drivers and the main feedbacks that could push the Amazon
forest towards a tipping point. From examples of disturbed forests
across the Amazon, we analyse the most plausible ecosystem trajec-
tories that may lead to alternative stable states10. Moreover, inspired
by the framework of ‘planetary boundaries’11, we identify climatic and
land use boundaries that reveal a safe operating space for the Amazon
forest system in the Anthropocene epoch12.
Theory and concepts
Over time, environmental conditions fluctuate and may cause stress
on ecosystems (for example, lack of water for plants). When stressing
conditions intensify, some ecosystems may change their equilibrium
state gradually, whereas others may shift abruptly between alternative
stable states10. A ‘tipping point’ is the critical threshold value of an
environmental stressing condition at which a small disturbance may
cause an abrupt shift in the ecosystem state2,3,13,14, accelerated by posi-
tive feedbacks15 (see Extended Data Table 1). This type of behaviour in
which the system gets into a phase of self-reinforcing (runaway) change
is often referred to as ‘critical transition’16. As ecosystems approach
a tipping point, they often lose resilience while still remaining close
to equilibrium17. Thus, monitoring changes in ecosystem resilience
and in key environmental conditions may enable societies to man-
age and avoid critical transitions. We adopt the concept of ‘ecological
resilience’18 (hereafter ‘resilience’), which refers to the ability of an
ecosystem to persist with similar structure, functioning and interac-
tions, despite disturbances that push it to an alternative stable state.
The possibility that alternative stable states (or bistability) may exist in
a system has important implications, because the crossing of tipping
points may be irreversible for the time scales that matter to societies10.
Tropical terrestrial ecosystems are a well-known case in which critical
transitions between alternative stable states may occur (Extended
Data Fig. 1).
Past dynamics
The Amazon system has been mostly covered by forest throughout
the Cenozoic era19 (for 65 million years). Seven million years ago, the
Amazon river began to drain the massive wetlands that covered most
of the western Amazon, allowing forests to expand over grasslands
in that region. More recently, during the drier and cooler conditions
of the Last Glacial Maximum20 (LGM) (around 21,000 years ago) and
of the mid-Holocene epoch21 (around 6,000 years ago), forests per-
sisted even when humans were already present in the landscape22.
Nonetheless, savannas expanded in peripheral parts of the southern
Amazon basin during the LGM and mid-Holocene23, as well as in the
northeastern Amazon during the early Holocene (around 11,000
years ago), probably influenced by drier climatic conditions and
fires ignited by humans24,25. Throughout the core of the Amazon for-
est biome, patches of white-sand savanna also expanded in the past
20,000–7,000 years, driven by sediment deposition along ancient
rivers26, and more recently (around 800 years ago) owing to Indigenous
fires27. However, during the past 3,000 years, forests have been mostly
expanding over savanna in the southern Amazon driven by increasingly
wet conditions28.
Although palaeorecords suggest that a large-scale Amazon forest col-
lapse did not occur within the past 65 million years19, they indicate that
savannas expanded locally, particularly in the more seasonal peripheral
556 | Nature | Vol 626 | 15 February 2024
regions when fires ignited by humans were frequent23,24. Patches of
white-sand savanna also expanded within the Amazon forest owing
to geomorphological dynamics and fires26,27. Past drought periods
were usually associated with much lower atmospheric CO2 concentra-
tions, which may have reduced water-use efficiency of trees29 (that is,
trees assimilated less carbon during transpiration). However, these
periods also coincided with cooler temperatures20,21, which probably
reduced water demand by trees30. Past drier climatic conditions were
therefore very different from the current climatic conditions, in which
observed warming trends may exacerbate drought impacts on the
forest by exposing trees to unprecedented levels of water stress31,32.
Global change impacts on forest resilience
Satellite observations from across the Amazon suggest that forest
resilience has been decreasing since the early 2000s33, possibly as a
result of global changes. In this section, we synthesize three global
change impacts that vary spatially and temporally across the Amazon
system, affecting forest resilience and the risk of critical transitions.
Regional climatic conditions
Within the twenty-first century, global warming may cause long-term
changes in Amazonian climatic conditions2. Human greenhouse gas
emissions continue to intensify global warming, but the warming rate
also depends on feedbacks in the climate system that remain uncertain2,3.
Recent climate models of the 6th phase of the Coupled Model Inter-
comparison Project (CMIP6) agree that in the coming decades, rainfall
conditions will become more seasonal in the eastern and southern
Amazonian regions, and temperatures will become higher across the
entire Amazon1,2. By 2050, models project that a significant increase
in the number of consecutive dry days by 10−30 days and in annual
maximum temperatures by 2–4 °C, depending on the greenhouse gas
emission scenario2. These climatic conditions could expose the forest
to unprecedented levels of vapour pressure deficit31 and consequently
water stress30.
Satellite observations of climatic variability31 confirm model projec-
tions2, showing that since the early 1980s, the Amazonian region has
been warming significantly at an average rate of 0.27 °C per decade
during the dry season, with the highest rates of up to 0.6 °C per decade
in the centre and southeast of the biome (Fig. 1a). Only a few small areas
in the west of the biome are significantly cooling by around 0.1 °C per
decade (Fig. 1a). Dry season mean temperature is now more than 2 °C
higher than it was 40 years ago in large parts of the central and south-
eastern Amazon. If trends continue, these areas could potentially warm
by over 4 °C by 2050. Maximum temperatures during the dry season
follow a similar trend, rising across most of the biome (Extended Data
Fig. 2), exposing the forest34 and local peoples35 to potentially unbear-
able heat. Rising temperatures will increase thermal stress, potentially
reducing forest productivity and carbon storage capacity36 and causing
widespread leaf damage34.
Since the early 1980s, rainfall conditions have also changed31.
Peripheral and central parts of the Amazon forest are drying signifi-
cantly, such as in the southern Bolivian Amazon, where annual rain-
fall reduced by up to 20 mm yr−1 (Extended Data Fig. 3a). By contrast,
parts of the western and eastern Amazon forest are becoming wetter,
with annual rainfall increasing by up to 20 mm yr−1. If these trends con-
tinue, ecosystem stability (as in Extended Data Fig. 1) will probably
change in parts of the Amazon by 2050, reshaping forest resilience to
disturbances (Fig. 1b and Extended Data Fig. 3b). For example, 6% of
the biome may change from stable forest to a bistable regime in parts
of the southern and central Amazon. Another 3% of the biome may
pass the critical threshold in annual rainfall into stable savanna in the
southern Bolivian Amazon. Bistable areas covering 8% of the biome
may turn into stable forest in the western Amazon (Peru and Bolivia),
thus becoming more resilient to disturbances. For comparison with
 Amazon biome
a b
Temperature
slope (ºC yr–1)
0.06
–0.01
km
0 250 500
d e
Roads
Forest fires
km
0 250 500
Fig. 1 | Exploring ecosystem transition potential across the Amazon forest
biome as a result of compounding disturbances. a, Changes in the dry season
(July–October) mean temperature reveal widespread warming, estimated
using simple regressions between time and temperature observed between
1981 and 2020 (with P < 0.1). b, Potential ecosystem stability classes estimated
for year 2050, adapted from current stability classes (Extended Data Fig. 1b) by
considering only areas with significant regression slopes between time and
annual rainfall observed from 1981 through 2020 (with P < 0.1) (see Extended
Data Fig. 3 for areas with significant changes). c, Repeated extreme drought
events between 2001–2018 (adapted from ref. 39). d, Road network from
where illegal deforestation and degradation may spread. e, Protected areas
and Indigenous territories reduce deforestation and fire disturbances.
f, Ecosystem transition potential (the possibility of forest shifting into an
satellite observations, we used projections of ecosystem stability by
2050 based on CMIP6 model ensembles for a low (SSP2–4.5) and a high
(SSP5–8.5) greenhouse gas emission scenario (Extended Data Fig. 4
and Supplementary Table 1). An ensemble with the 5 coupled models
that include a dynamic vegetation module indicates that 18–27% of
the biome may transition from stable forest to bistable and that 2–6%
may transition to stable savanna (depending on the scenario), mostly
in the northeastern Amazon. However, an ensemble with all 33 models
suggests that 35–41% of the biome could become bistable, including
large areas of the southern Amazon. The difference between both
ensembles is possibly related to the forest–rainfall feedback included
in the five coupled models, which increases total annual rainfall and
therefore the stable forest area along the southern Amazon, but only
when deforestation is not included in the simulations4,37. Nonethe-
less, both model ensembles agree that bistable regions will expand
deeper into the Amazon, increasing the risk of critical transitions due
to disturbances (as implied by the existence of alternative stable states;
Extended Data Fig. 1).
Disturbance regimes
Within the remaining Amazon forest area, 17% has been degraded by
human disturbances38, such as logging, edge effects and understory
fires, but if we consider also the impacts from repeated extreme drought
Countries Amazon basin
c
Stable forest Number of extreme
Bistable drought events
Stable savanna 5
1
km km
0 250 500 0 250 500
By year 2050
f
Indigenous territories Tipping potential
Protected areas 4
–1
km km
0 250 500 0 250 500
alternative structural or compositional state) across the Amazon biome by year
2050 inferred from compounding disturbances (a–d) and high-governance
areas (e). We excluded accumulated deforestation until 2020 and savannas.
Transition potential rises with compounding disturbances and varies as
follows: less than 0 (in blue) as low; between 1 and 2 as moderate (in yellow);
more than 2 as high (orange–red). Transition potential represents the sum
of: (1) slopes of dry season mean temperature (as in a, multiplied by 10);
(2) ecosystem stability classes estimated for year 2050 (as in b), with 0 for
stable forest, 1 for bistable and 2 for stable savanna; (3) accumulated impacts
from extreme drought events, with 0.2 for each event; (4) road proximity as
proxy for degrading activities, with 1 for pixels within 10 km from a road;
(5) areas with higher governance within protected areas and Indigenous
territories, with −1 for pixels inside these areas. For more details, see Methods.
events in the past decades, 38% of the Amazon could be degraded39.
Increasing rainfall variability is causing extreme drought events to
become more widespread and frequent across the Amazon (Fig. 1c),
together with extreme wet events and convective storms that result
in more windthrow disturbances40. Drought regimes are intensifying
across the region41, possibly due to deforestation42 that continues to
expand within the system (Extended Data Fig. 5). As a result, new fire
regimes are burning larger forest areas43, emitting more carbon to the
atmosphere44 and forcing IPLCs to readapt45. Road networks (Fig. 1d)
facilitate illegal activities, promoting more deforestation, logging and
fire spread throughout the core of the Amazon forest38,39. The impacts
of these pervasive disturbances on biodiversity and on IPLCs will prob-
ably affect ecosystem adaptability (Box 1), and consequently forest
resilience to global changes.
Currently, 86% of the Amazon biome may be in a stable forest state
(Extended Data Fig. 1b), but some of these stable forests are showing
signs of fragility33. For instance, field evidence from long-term moni-
toring sites across the Amazon shows that tree mortality rates are
increasing in most sites, reducing carbon storage46, while favouring
the replacement by drought-affiliated species47. Aircraft measure-
ments of vertical carbon flux between the forest and atmosphere reveal
how southeastern forests are already emitting more carbon than they
absorb, probably because of deforestation and fire48.
Nature | Vol 626 | 15 February 2024 | 557
Analysis
Box 1
Ecosystem adaptability
We define ‘ecosystem adaptability’ as the capacity of an ecosystem
to reorganize and persist in the face of environmental changes.
In the past, many internal mechanisms have probably contributed
to ecosystem adaptability, allowing Amazonian forests to persist
during times of climate change. In this section we synthesize two
of these internal mechanisms, which are now being undermined by
global change.
Biodiversity
Amazonian forests are home to more than 15,000 tree species,
of which 1% are dominant and the other 99% are mostly rare107.
A single forest hectare in the central and northwestern Amazon can
contain more than 300 tree species (Extended Data Fig. 7a). Such
tremendous tree species diversity can increase forest resilience
by different mechanisms. Tree species complementarity increases
carbon storage, accelerating forest recovery after disturbances108.
Tree functional diversity increases forest adaptability to climate
chance by offering various possibilities of functioning99. Rare
species provide ‘ecological redundancy’, increasing opportunities
for replacement of lost functions when dominant species
disappear109. Diverse forests are also more likely to resist severe
disturbances owing to ‘response diversity’110—that is, some species
may die, while others persist. For instance, in the rainy western
Amazon, drought-resistant species are rare but present within tree
communities111, implying that they could replace the dominant
drought-sensitive species in a drier future. Diversity of other
organisms, such as frugivores and pollinators, also increases forest
resilience by stabilizing ecological networks15,112. Considering that
half of Amazonian tree species are estimated to become threatened
As bistable forests expand deeper into the system (Fig. 1b and
Extended Data Fig. 4), the distribution of compounding disturbances
may indicate where ecosystem transitions are more likely to occur in
the coming decades (Fig. 1f). For this, we combined spatial informa-
tion on warming and drying trends, repeated extreme drought events,
together with road networks, as proxy for future deforestation and
degradation38,39. We also included protected areas and Indigenous
territories as areas with high forest governance, where deforestation
and fire regimes are among the lowest within the Amazon49 (Fig. 1e).
This simple additive approach does not consider synergies between
compounding disturbances that could trigger unexpected ecosystem
transitions. However, by exploring only these factors affecting forest
resilience and simplifying the enormous Amazonian complexity, we
aimed to produce a simple and comprehensive map that can be use-
ful for guiding future governance. We found that 10% of the Amazon
forest biome has a relatively high transition potential (more than 2
disturbance types; Fig. 1f), including bistable forests that could tran-
sition into a low tree cover state near savannas of Guyana, Venezuela,
Colombia and Peru, as well as stable forests that could transition into
alternative compositional states within the central Amazon, such as
along the BR319 and Trans-Amazonian highways. Smaller areas with
high transition potential were found scattered within deforestation
frontiers, where most forests have been carved by roads50,51. Moreo-
ver, 47% of the biome has a moderate transition potential (more than
1 disturbance type; Fig. 1f), including relatively remote parts of the
central Amazon where warming trends and repeated extreme drought
events overlap (Fig. 1a,c). By contrast, large remote areas covering
53% of the biome have low transition potential, mostly reflecting the
558 | Nature | Vol 626 | 15 February 2024
(IUCN Red list) by 2050 owing to climate change, deforestation
and degradation8, biodiversity losses could contribute to further
reducing forest resilience.
Indigenous peoples and local communities
Globally, Indigenous peoples and local communities (IPLCs) have
a key role in maintaining ecosystems resilient to global change113.
Humans have been present in the Amazon for at least 12,000 years114
and extensively managing landscapes for 6,000 years22. Through
diverse ecosystem management practices, humans built thousands
of earthworks and ‘Amazon Dark Earth’ sites, and domesticated
plants and landscapes across the Amazon forest115,116. By creating
new cultural niches, humans partly modified the Amazonian flora117,118,
increasing their food security even during times of past climate
change119,120 without the need for large-scale deforestation117.
Today, IPLCs have diverse ecological knowledge about Amazonian
plants, animals and landscapes, which allows them to quickly
identify and respond to environmental changes with mitigation and
adaptation practices68,69. IPLCs defend their territories against illegal
deforestation and land use disturbances49,113, and they also promote
forest restoration by expanding diverse agroforestry systems121,122.
Amazonian regions with the highest linguistic diversity (a proxy for
ecological knowledge diversity123) are found in peripheral parts of
the system, particularly in the north-west (Extended Data Fig. 7b).
However, consistent loss of Amazonian languages is causing an
irreversible disruption of ecological knowledge systems, mostly driven
by road construction7. Continued loss of ecological knowledge will
undermine the capacity of IPLCs to manage and protect Amazonian
forests, further reducing their resilience to global changes9.
distribution of protected areas and Indigenous territories (Fig. 1e). If
these estimates, however, considered projections from CMIP6 models
and their relatively broader areas of bistability (Extended Data Fig. 4),
the proportion of the Amazon forest that could transition into a low
tree cover state would be much larger.
CO2 fertilization
Rising atmospheric CO2 concentrations are expected to increase the
photosynthetic rates of trees, accelerating forest growth and bio-
mass accumulation on a global scale52. In addition, CO2 may reduce
water stress by increasing tree water-use efficiency29. As result, a ‘CO2
fertilization effect’ could increase forest resilience to climatic vari-
ability53,54. However, observations from across the Amazon46 suggest
that CO2-driven accelerations of tree growth may have contributed to
increasing tree mortality rates (trees grow faster but also die earlier),
which could eventually neutralize the forest carbon sink in the com-
ing decades55. Moreover, increases in tree water-use efficiency may
reduce forest transpiration and consequently atmospheric moisture
flow across the Amazon53,56, potentially reducing forest resilience in
the southwest of the biome4,37. Experimental evidence suggests that
CO2 fertilization also depends on soil nutrient availability, particularly
nitrogen and phosphorus57,58. Thus, it is possible that in the fertile soils
of the western Amazon and Várzea floodplains, forests may gain resil-
ience from increasing atmospheric CO2 (depending on how it affects
tree mortality rates), whereas on the weathered (nutrient-poor) soils
across most of the Amazon basin59, forests might not respond to atmos-
pheric CO2 increase, particularly on eroded soils within deforestation
frontiers60. In sum, owing to multiple interacting factors, potential
responses of Amazonian forests to CO2 fertilization are still poorly
understood. Forest responses depend on scale, with resilience possibly
increasing at the local scale on relatively more fertile soils, but decreas-
ing at the regional scale due to reduced atmospheric moisture flow.
Local versus systemic transition
Environmental heterogeneity
Environmental heterogeneity can reduce the risk of systemic transi-
tion (large-scale forest collapse) because when stressing conditions
intensify (for example, rainfall declines), heterogeneous forests may
transition gradually (first the less resilient forest patches, followed by
the more resilient ones), compared to homogeneous forests that may
transition more abruptly17 (all forests transition in synchrony). Ama-
zonian forests are heterogeneous in their resilience to disturbances,
which may have contributed to buffering large-scale transitions in the
past37,61,62. At the regional scale, a fundamental heterogeneity factor is
rainfall and how it translates into water stress. Northwestern forests
rarely experience water stress, which makes them relatively more resil-
ient than southeastern forests that may experience water stress in the
dry season, and therefore are more likely to shift into a low tree cover
state. As a result of low exposure to water deficit, most northwestern
forests have trees with low drought resistance and could suffer mas-
sive mortality if suddenly exposed to severe water stress32. However,
this scenario seems unlikely to occur in the near future (Fig. 1). By
contrast, most seasonal forest trees have various strategies to cope
with water deficit owing to evolutionary and adaptive responses to
historical drought events32,63. These strategies may allow seasonal
forests to resist current levels of rainfall fluctuations32, but seasonal
forests are also closer to the critical rainfall thresholds (Extended
Data Fig. 1) and may experience unprecedented water stress in the
coming decades (Fig. 1).
Other key heterogeneity factors (Extended Data Fig. 6) include
topography, which determines plant access to groundwater64, and
seasonal flooding, which increases forest vulnerability to wildfires65.
Future changes in rainfall regimes will probably affect hydrological
regimes66, exposing plateau (hilltop) forests to unprecedented water
stress, and floodplain forests to extended floods, droughts and wild-
fires. Soil fertility is another heterogeneity factor that may affect for-
est resilience59, and which may be undermined by disturbances that
cause topsoil erosion60. Moreover, as human disturbances intensify
throughout the Amazon (Fig. 1), the spread of invasive grasses and fires
can make the system increasingly homogeneous. Effects of heterogene-
ity on Amazon forest resilience have been poorly investigated so far
(but see refs. 37,61,62) and many questions remain open, such as how
much heterogeneity exists in the system and whether it can mitigate
a systemic transition.
Sources of connectivity
Connectivity across Amazonian landscapes and regions can contribute
to synchronize forest dynamics, causing different forests to behave
more similarly17. Depending on the processes involved, connectivity
can either increase or decrease the risk of systemic transition17. For
instance, connectivity may facilitate forest recovery after disturbances
through seed dispersal, but also it may spread disturbances, such as
fire. In the Amazon, an important source of connectivity enhancing
forest resilience is atmospheric moisture flow westward (Fig. 2), partly
maintained by forest evapotranspiration4,37,67. Another example of con-
nectivity that may increase social-ecological resilience is knowledge
exchange among IPLCs about how to adapt to global change68,69 (see
Box 1). However, complex systems such as the Amazon can be particu-
larly vulnerable to sources of connectivity that spread disturbances and
increase the risk of systemic transition70. For instance, roads carving
through the forest are well-known sources of illegal activities, such as
logging and burning, which increase forest flammability38,39.
Amazon biome Countries Amazon basin Moisture flow
Guyana
Venezuela
Suriname
French
10%
Guyana
Colombia
13%
Ecuador
19%
16%
21% Brazil
Peru
33%
10%
Bolivia
0
km
250 500
Fig. 2 | Connectivity between Amazonian countries involving atmospheric
moisture flow. Brazil holds 60% of the Amazon forest biome and has a major
responsibility towards its neighbouring countries in the west. Brazil is the
largest supplier of rainfall to western Amazonian countries. Up to one-third of
the total annual rainfall in Amazonian territories of Bolivia, Peru, Colombia and
Ecuador depends on water originating from Brazil’s portion of the Amazon
forest. This international connectivity illustrates how policies related to
deforestation, especially in the Brazilian Amazon, will affect the climate in
other countries. Arrow widths are proportional to the percentage of the annual
rainfall received by each country within their Amazonian areas. We only show
flows with percentages higher than 10% (see Methods for details).
Five critical drivers of water stress
Global warming
Most CMIP6 models agree that a large-scale dieback of the Amazon
is unlikely in response to global warming above pre-industrial levels2,
but this ecosystem response is based on certain assumptions, such as
a large CO2-fertilization effect53. Forests across the Amazon are already
responding with increasing tree mortality rates that are not simulated
by these models46, possibly because of compounding disturbance
regimes (Fig. 1). Nonetheless, a few global climate models3,14,71–74 indi-
cate a broad range for a potential critical threshold in global warming
between 2 and 6 °C (Fig. 3a). These contrasting results can be explained
by general differences between numerical models and their represen-
tation of the complex Amazonian system. While some models with
dynamic vegetation indicate local-scale tipping events in peripheral
parts of the Amazon5,6, other models suggest an increase in biomass
and forest cover (for example, in refs. 53,54). For instance, a study found
that when considering only climatic variability, a large-scale Amazon
forest dieback is unlikely, even under a high greenhouse gas emission
scenario75. However, most updated CMIP6 models agree that droughts
in the Amazon region will increase in length and intensity, and that
exceptionally hot droughts will become more common2, creating con-
ditions that will probably boost other types of disturbances, such as
large and destructive forest fires76,77. To avoid broad-scale ecosystem
transitions due to synergies between climatic and land use disturbances
(Fig. 3b), we suggest a safe boundary for the Amazon forest at 1.5 °C for
global warming above pre-industrial levels, in concert with the Paris
Agreement goals.
Annual rainfall
Satellite observations of tree cover distributions across tropical South
America suggest a critical threshold between 1,000 and 1,250 mm of
Nature | Vol 626 | 15 February 2024 | 559
Analysis
a b
Drivers of water stress Critical threshold
2 ºC
1. Global warming
(2 to 6 ºC)
1,000 mm
2. Annual rainfall
(800 to 1,150 mm)
–450 mm
3. Rainfall seasonality
(–400 to –500 mm)
intensity (MCWD)
8 months
4. Dry season length
(7.5 to 8.5 months)
20% deforested
5. Accumulated deforestation
(20 to 50%)
Fig. 3 | Drivers of water stress on the Amazon forest, their critical thresholds,
safe boundaries and interactions. a, Five critical drivers of water stress on
Amazonian forests affect (directly or indirectly) the underlying tipping point of
the system. For each driver, we indicate potential critical thresholds and safe
boundaries that define a safe operating space for keeping the Amazon forest
resilient11,12. We followed the precautionary principle and considered the
most conservative thresholds within the ranges, when confidence was low.
b, Conceptual model showing how the five drivers may interact (arrows indicate
positive effects) and how these interactions may strengthen a positive feedback
between water stress and forest loss. These emerging positive feedback loops
could accelerate a systemic transition of the Amazon forest15. At global scales,
annual rainfall78,79. On the basis of our reanalysis using tree cover data
from the Amazon basin (Extended Data Fig. 1a), we confirm a potential
threshold at 1,000 mm of annual rainfall (Fig. 3a), below which forests
become rare and unstable. Between 1,000 and 1,800 mm of annual
rainfall, high and low tree cover ecosystems exist in the Amazon as two
alternative stable states (see Extended Data Table 2 for uncertainty
ranges). Within the bistability range in annual rainfall conditions, for-
ests are relatively more likely to collapse when severely disturbed,
when compared to forests in areas with annual rainfall above 1,800 mm
(Extended Data Fig. 1a). For floodplain ecosystems covering 14% of
the forest biome, a different critical threshold has been estimated at
1,500 mm of annual rainfall65, implying that floodplain forests may
be the first to collapse in a drier future. To avoid local-scale ecosys-
tem transitions due to compounding disturbances, we suggest a safe
boundary in annual rainfall conditions at 1,800 mm.
Rainfall seasonality intensity
Satellite observations of tree cover distributions across tropical South
America suggest a critical threshold in rainfall seasonality intensity at
−400 mm of the maximum cumulative water deficit37,80 (MCWD). Our
reanalysis of the Amazon basin (Extended Data Fig. 1c) confirms the
critical threshold at approximately −450 mm in the MCWD (Fig. 3a), and
suggests a bistability range between approximately −350 and −450 mm
(see Extended Data Table 2 for uncertainty ranges), in which forests are
more likely to collapse when severely disturbed than forests in areas
with MCWD below −350 mm. To avoid local-scale ecosystem transi-
tions due to compounding disturbances, we suggest a safe boundary
of MCWD at −350 mm.
Dry season length
Satellite observations of tree cover distributions across tropical South
America suggest a critical threshold at 7 months of dry season length79
(DSL). Our reanalysis of the Amazon basin (Extended Data Fig. 1d)
suggests a critical threshold at eight months of DSL (Fig. 3a), with a
bistability range between approximately five and eight months (see
Extended Data Table 2 for uncertainty ranges), in which forests are
more likely to collapse when severely disturbed than forests in areas
with DSL below five months. To avoid local-scale ecosystem transitions
due to compounding disturbances, we suggest a safe boundary of DSL
at five months.
560 | Nature | Vol 626 | 15 February 2024
1. Global warming
Safe boundary Confidence
1.5 ºC Low3,14,71–74 2–4. Regional rainfall conditions
1,800 mm Medium78,79
Water
–350 mm Medium37,80
stress
+
Forest
5 months Medium79
loss
10% deforested Low67,81
5. Accumulated deforestation
driver 1 (global warming) intensifies with greenhouse gas emissions, including
emissions from deforestation. At local scales, driver 5 (accumulated deforestation)
intensifies with land use changes. Drivers 2 to 4 (regional rainfall conditions)
intensify in response to drivers 1 and 5. The intensification of these drivers may
cause widespread tree mortality for instance because of extreme droughts and
fires76. Water stress affects vegetation resilience globally79,104, but other stressors,
such as heat stress34,36, may also have a role. In the coming decades, these five
drivers could change at different rates, with some approaching a critical
threshold faster than others. Therefore, monitoring them separately can
provide vital information to guide mitigation and adaptation strategies.
Accumulated deforestation
A potential vegetation model81 found a critical threshold at 20% of accu-
mulated deforestation (Fig. 3a) by simulating Amazon forest responses
to different scenarios of accumulated deforestation (with associated
fire events) and of greenhouse gas emissions, and by considering a CO2
fertilization effect of 25% of the maximum photosynthetic assimilation
rate. Beyond 20% deforestation, forest mortality accelerated, causing
large reductions in regional rainfall and consequently an ecosystem
transition of 50−60% of the Amazon, depending on the emissions
scenario. Another study using a climate-vegetation model found that
with accumulated deforestation of 30−50%, rainfall in non-deforested
areas downwind would decline67 by 40% (ref. 67), potentially causing
more forest loss4,37. Other more recent models incorporating fire dis-
turbances support a potential broad-scale transition of the Amazon
forest, simulating a biomass loss of 30–40% under a high-emission
scenario5,82 (SSP5–8.5 at 4 °C). The Amazon biome has already lost 13%
of its original forest area due to deforestation83 (or 15% of the biome
if we consider also young secondary forests83 that provide limited
contribution to moisture flow84). Among the remaining old-growth
forests, at least 38% have been degraded by land use disturbances and
repeated extreme droughts39, with impacts on moisture recycling that
are still uncertain. Therefore, to avoid broad-scale ecosystem transi-
tions due to runaway forest loss (Fig. 3b), we suggest a safe boundary
of accumulated deforestation of 10% of the original forest biome cover,
which requires ending large-scale deforestation and restoring at least
5% of the biome.
Three alternative ecosystem trajectories
Degraded forest
In stable forest regions of the Amazon with annual rainfall above
1,800 mm (Extended Data Fig. 1b), forest cover usually recovers within
a few years or decades after disturbances, yet forest composition and
functioning may remain degraded for decades or centuries84–87. Esti-
mates from across the Amazon indicate that approximately 30% of
areas previously deforested are in a secondary forest state83 (covering
4% of the biome). An additional 38% of the forest biome has been dam-
aged by extreme droughts, fires, logging and edge effects38,39. These
forests may naturally regrow through forest succession, yet because of
 White-sand savanna
Disturbances
Extreme droughts
Fires
Low tree cover
Feedbacks
Seed Fires
limitation
Soil erosion
Alternative states
Low tree cover
Fig. 4 | Alternative ecosystem trajectories for Amazonian forests that
transition due to compounding disturbances. From examples of disturbed
forests across the Amazon, we identify the three most plausible ecosystem
trajectories related to the types of disturbances, feedbacks and local
environmental conditions. These alternative trajectories may be irreversible
or transient depending on the strength of the novel interactions15. Particular
combinations of interactions (arrows show positive effects described in the
literature) may form feedback loops15 that propel the ecosystem through
these trajectories. In the ‘degraded forest’ trajectory, feedbacks often involve
competition between trees and other opportunistic plants85,90,92, as well as
interactions between deforestation, fire and seed limitation84,87,105. At the
landscape scale, secondary forests are more likely to be cleared than mature
forests, thus keeping forests persistently young and landscapes fragmented83.
feedbacks15, succession can become arrested, keeping forests persis-
tently degraded (Fig. 4). Different types of degraded forests have been
identified in the Amazon, each one associated with a particular group
of dominant opportunistic plants. For instance, Vismia forests are com-
mon in old abandoned pastures managed with fire85, and are relatively
stable, because Vismia trees favour recruitment of Vismia seedlings in
detriment of other tree species88,89. Liana forests can also be relatively
stable, because lianas self-perpetuate by causing physical damage
to trees, allowing lianas to remain at high density90,91. Liana forests
are expected to expand with increasing aridity, disturbance regimes
and CO2 fertilization90. Guadua bamboo forests are common in the
southwestern Amazon92,93. Similar to lianas, bamboos self-perpetuate
by causing physical damage to trees and have been expanding over
burnt forests in the region92. Degraded forests are usually dominated
by native opportunistic species, and their increasing expansion over
disturbed forests could affect Amazonian functioning and resilience
in the future.
White-sand savanna
White-sand savannas are ancient ecosystems that occur in patches
within the Amazon forest biome, particularly in seasonally waterlogged
or flooded areas94. Their origin has been attributed to geomorphologi-
cal dynamics and past Indigenous fires26,27,94. In a remote landscape far
from large agricultural frontiers, within a stable forest region of the
Amazon (Extended Data Fig. 1b), satellite and field evidence revealed
that white-sand savannas are expanding where floodplain forests were
Degraded open canopy Degraded forest
Warming Warming
Bistability Extreme droughts
Extreme droughts Deforestation
Fires Fires
Low tree cover Degraded
forest
Alien Opportunistic
Seed grasses Fires Deforestation
plants
limitation
Soil erosion Seed Fires
limitation
High tree cover
In the ‘degraded open-canopy ecosystem’ trajectory, feedbacks involve
interactions among low tree cover and fire97, soil erosion60, seed limitation105,
invasive grasses and opportunistic plants96. At the regional scale, a self-reinforcing
feedback between forest loss and reduced atmospheric moisture flow may
increase the resilience of these open-canopy degraded ecosystems42. In the
‘white-sand savanna’ trajectory, the main feedbacks result from interactions
among low tree cover and fire, soil erosion, and seed limitation106. Bottom left,
floodplain forest transition to white-sand savanna after repeated fires (photo
credit: Bernardo Flores); bottom centre, forest transition to degraded open-
canopy ecosystem after repeated fires (photo credit: Paulo Brando); bottom
right, forest transition to Vismia degraded forest after slash-and-burn
agriculture (photo credit: Catarina Jakovac).
repeatedly disturbed by fires95. After fire, the topsoil of burnt forests
changes from clayey to sandy, favouring the establishment of savanna
trees and native herbaceous plants95. Shifts from forest to white-sand
savanna (Fig. 4) are probably stable (that is, the ecosystem is unlikely
to recover back to forest within centuries), based on the relatively long
persistence of these savannas in the landscape94. Although these eco-
system transitions have been confirmed only in the Negro river basin
(central Amazon), floodplain forests in other parts of the Amazon were
shown to be particularly vulnerable to collapse45,64,65.
Degraded open-canopy ecosystem
In bistable regions of the Amazon forest with annual rainfall below
1,800 mm (Extended Data Fig. 1b), shifts to degraded open-canopy
ecosystems are relatively common after repeated disturbances by
fire45,96. The ecosystem often becomes dominated by fire-tolerant tree
and palm species, together with alien invasive grasses and opportun-
istic herbaceous plants96,97, such as vines and ferns. Estimates from
the southern Amazon indicate that 5−6% of the landscape has already
shifted into degraded open-canopy ecosystems due to deforestation
and fires45,96. It is still unclear, however, whether degraded open-canopy
ecosystems are stable or transient (Fig. 4). Palaeorecords from the
northern Amazon98 show that burnt forests may spend centuries
in a degraded open-canopy state before they eventually shift into a
savanna. Today, invasion by alien flammable grasses is a novel stabiliz-
ing mechanism96,97, but the long-term persistence of these grasses in
the ecosystem is also uncertain.
Nature | Vol 626 | 15 February 2024 | 561
Analysis
Prospects for modelling Amazon forest dynamics
Several aspects of the Amazon forest system may help improve earth
system models (ESMs) to more accurately simulate ecosystem dynam-
ics and feedbacks with the climate system. Simulating individual trees
can improve the representation of growth and mortality dynamics,
which ultimately affect forest dynamics (for example, refs. 61,62,99).
Significant effects on simulation results may emerge from increas-
ing plant functional diversity, representation of key physiological
trade-offs and other features that determine water stress on plants,
and also allowing for community adjustment to environmental hetero-
geneity and global change32,55,62,99. For now, most ESMs do not simulate
a dynamic vegetation cover (Supplementary Table 1) and biomes are
represented based on few plant functional types, basically simulating
monocultures on the biome level. In reality, tree community adap-
tation to a heterogenous and dynamic environment feeds into the
whole-system dynamics, and not covering such aspects makes a true
Amazon tipping assessment more challenging.
Our findings also indicate that Amazon forest resilience is affected
by compounding disturbances (Fig. 1). ESMs need to include differ-
ent disturbance scenarios and potential synergies for creating more
realistic patterns of disturbance regimes. For instance, logging and
edge effects can make a forest patch more flammable39, but these dis-
turbances are often not captured by ESMs. Improvements in the ability
of ESMs to predict future climatic conditions are also required. One way
is to identify emergent constraints100, lowering ESMs variations in their
projections of the Amazonian climate. Also, fully coupled ESMs simu-
lations are needed to allow estimates of land-atmosphere feedbacks,
which may adjust climatic and ecosystem responses. Another way to
improve our understanding of the critical thresholds for Amazonian
resilience and how these link to climatic conditions and to greenhouse
gas concentrations is through factorial simulations with ESMs. In sum,
although our study may not deliver a set of reliable and comprehen-
sive equations to parameterize processes impacting Amazon forest
dynamics, required for implementation in ESMs, we highlight many
of the missing modelled processes.
Implications for governance
Forest resilience is changing across the Amazon as disturbance regimes
intensify (Fig. 1). Although most recent models agree that a large-scale
collapse of the Amazon forest is unlikely within the twenty-first cen-
tury2, our findings suggest that interactions and synergies among dif-
ferent disturbances (for example, frequent extreme hot droughts and
forest fires) could trigger unexpected ecosystem transitions even in
remote and central parts of the system101. In 2012, Davidson et al.102
demonstrated how the Amazon basin was experiencing a transition
to a ‘disturbance-dominated regime’ related to climatic and land use
changes, even though at the time, annual deforestation rates were
declining owing to new forms of governance103. Recent policy and
approaches to Amazon development, however, accelerated deforesta-
tion that reached 13,000 km2 in the Brazilian Amazon in 2021 (http://
terrabrasilis.dpi.inpe.br). The southeastern region has already turned
into a source of greenhouse gases to the atmosphere48. The conse-
quences of losing the Amazon forest, or even parts of it, imply that we
must follow a precautionary approach—that is, we must take actions
that contribute to maintain the Amazon forest within safe boundaries12.
Keeping the Amazon forest resilient depends firstly on humanity’s
ability to stop greenhouse gas emissions, mitigating the impacts of
global warming on regional climatic conditions2. At the local scale,
two practical and effective actions need to be addressed to reinforce
forest–rainfall feedbacks that are crucial for the resilience of the
Amazon forest4,37: (1) ending deforestation and forest degradation;
and (2) promoting forest restoration in degraded areas. Expanding
protected areas and Indigenous territories can largely contribute to
562 | Nature | Vol 626 | 15 February 2024
these actions. Our findings suggest a list of thresholds, disturbances
and feedbacks that, if well managed, can help maintain the Amazon
forest within a safe operating space for future generations.
Online content
Any methods, additional references, Nature Portfolio reporting summa-
ries, source data, extended data, supplementary information, acknowl-
edgements, peer review information; details of author contributions
and competing interests; and statements of data and code availability
are available at https://doi.org/10.1038/s41586-023-06970-0.
1. Science Panel for the Amazon. Amazon Assessment Report 2021 (2021);
www.theamazonwewant.org/amazon-assessment-report-2021/.
2. IPCC. Climate Change 2021: The Physical Science Basis (eds Masson-Delmotte, V. et al.)
https://www.ipcc.ch/report/ar6/wg1/#FullReport (Cambridge Univ. Press, 2021).
3. Armstrong McKay, D. et al. Exceeding 1.5 °C global warming could trigger multiple
climate tipping points. Science 377, abn7950 (2022).
4. Staal, A. et al. Forest-rainfall cascades buffer against drought across the Amazon.
Nat. Clim. Change 8, 539–543 (2018).
5. Cano, I. M. et al. Abrupt loss and uncertain recovery from fires of Amazon forests under
low climate mitigation scenarios. Proc. Natl Acad. Sci. USA 119, e2203200119 (2022).
6. Parry, I. M., Ritchie, P. D. L. & Cox, P. M. Evidence of localised Amazon rainforest dieback in
CMIP6 models. Earth Syst. Dynam. 13, 1667–1675 (2022).
7. Bromham, L. et al. Global predictors of language endangerment and the future of
linguistic diversity. Nat. Ecol. Evol. 6, 163–173 (2022).
8. Gomes, V. H. F., Vieira, I. C. G., Salomão, R. P. & ter Steege, H. Amazonian tree species
threatened by deforestation and climate change. Nat. Clim. Change 9, 547–553 (2019).
9. Cámara-Leret, R., Fortuna, M. A. & Bascompte, J. Indigenous knowledge networks in the
face of global change. Proc. Natl Acad. Sci. USA 116, 9913–9918 (2019).
10. Scheffer, M., Carpenter, S., Foley, J. A., Folke, C. & Walker, B. Catastrophic shifts in
ecosystems. Nature 413, 591–596 (2001).
11. Rockstrom, J. et al. A safe operating space for humanity. Nature 461, 472–475 (2009).
12. Scheffer, M. et al. Creating a safe operating space for iconic ecosystems. Science 347,
1317–1319 (2015).
13. van Nes, E. H. et al. What do you mean, ‘tipping point’? Trends Ecol. Evol. 31, 902–904
(2016).
14. Lenton, T. M. et al. Tipping elements in the Earth’s climate system. Proc. Natl Acad. Sci.
USA 105, 1786–1793 (2008).
15. Flores, B. M. & Staal, A. Feedback in tropical forests of the Anthropocene. Global Change
Biol. 28, 5041–5061 (2022).
16. Scheffer, M. Critical Transitions in Nature and Society (Princeton Univ. Press, 2009).
17. Scheffer, M. et al. Anticipating critical transitions. Science 338, 344–348 (2012).
18. Holling, C. S. Engineering Resilience versus Ecological Resilience (National Academy
Press, 1996).
19. Hoorn, C. et al. Amazonia through time: Andean uplift, climate change, landscape
evolution, and biodiversity. Science 330, 927–931 (2010).
20. Wang, X. et al. Hydroclimate changes across the Amazon lowlands over the past 45,000
years. Nature 541, 204–207 (2017).
21. Kukla, T. et al. The resilience of Amazon tree cover to past and present drying. Global
Planet. Change 202, 103520 (2021).
22. Clement, C. R. et al. Disentangling domestication from food production systems in the
neotropics. Quaternary 4, 4 (2021).
23. Mayle, F. E. & Power, M. J. Impact of a drier Early–Mid-Holocene climate upon Amazonian
forests. Phil. Trans. R. Soc. B 363, 1829–1838 (2008).
24. Montoya, E. & Rull, V. Gran Sabana fires (SE Venezuela): a paleoecological perspective.
Quat. Sci. Rev. 30, 3430–3444 (2011).
25. Rull, V., Montoya, E., Vegas-Vilarrúbia, T. & Ballesteros, T. New insights on palaeofires and
savannisation in northern South America. Quat. Sci. Rev. 122, 158–165 (2015).
26. Rossetti, D. F. et al. Unfolding long-term Late Pleistocene-Holocene disturbances of
forest communities in the southwestern Amazonian lowlands. Ecosphere 9, e02457
(2018).
27. Prance, G. T. & Schubart, H. O. R. Notes on the vegetation of Amazonia I. A preliminary
note on the origin of the open white sand campinas of the lower Rio Negro. Brittonia 30,
60 (1978).
28. Wright, J. L. et al. Sixteen hundred years of increasing tree cover prior to modern
deforestation in Southern Amazon and central Brazilian savannas. Glob. Change Biol. 27,
136–150 (2021).
29. van der Sleen, P. et al. No growth stimulation of tropical trees by 150 years of CO2
fertilization but water-use efficiency increased. Nat. Geosci. 8, 24–28 (2015).
30. Smith, M. N. et al. Empirical evidence for resilience of tropical forest photosynthesis in a
warmer world. Nat. Plants 6, 1225–1230 (2020).
31. Marengo, J. A., Jimenez, J. C., Espinoza, J.-C., Cunha, A. P. & Aragão, L. E. O. Increased
climate pressure on the agricultural frontier in the Eastern Amazonia–Cerrado transition
zone. Sci. Rep. 12, 457 (2022).
32. Tavares, J. V. et al. Basin-wide variation in tree hydraulic safety margins predicts the
carbon balance of Amazon forests. Nature 617, 111–117 (2023).
33. Boulton, C. A., Lenton, T. M. & Boers, N. Pronounced loss of Amazon rainforest resilience
since the early 2000s. Nat. Clim. Change 12, 271–278 (2022).
34. Doughty, C. E. et al. Tropical forests are approaching critical temperature thresholds.
Nature 621, 105–111 (2023).
35. Xu, C., Kohler, T. A., Lenton, T. M., Svenning, J.-C. & Scheffer, M. Future of the human
climate niche. Proc. Natl Acad. Sci. USA 117, 11350–11355 (2020).
36. Sullivan, M. J. P. et al. Long-term thermal sensitivity of Earth’s tropical forests. Science
368, 869–874 (2020).
37. Zemp, D. C. et al. Self-amplified Amazon forest loss due to vegetation-atmosphere
feedbacks. Nat. Commun. 8, 14681 (2017).
38. Bullock, E. L., Woodcock, C. E., Souza, C. Jr & Olofsson, P. Satellite-based estimates reveal
widespread forest degradation in the Amazon. Global Change Biol. 26, 2956–2969
(2020).
39. Lapola, D. M. et al. The drivers and impacts of Amazon forest degradation. Science 379,
eabp8622 (2023).
40. Feng, Y., Negrón-Juárez, R. I., Romps, D. M. & Chambers, J. Q. Amazon windthrow
disturbances are likely to increase with storm frequency under global warming. Nat.
Commun. 14, 101 (2023).
41. Anderson, L. O. et al. Vulnerability of Amazonian forests to repeated droughts. Phil. Trans.
R. Soc. B 373, 20170411 (2018).
42. Staal, A. et al. Feedback between drought and deforestation in the Amazon. Environ. Res.
Lett. 15, 044024 (2020).
43. Alencar, A. A., Brando, P. M., Asner, G. P. & Putz, F. E. Landscape fragmentation,
severe drought, and the new Amazon forest fire regime. Ecol. Appl. 25, 1493–1505
(2015).
44. Aragão, L. E. O. C. et al. 21st century drought-related fires counteract the decline of
Amazon deforestation carbon emissions. Nat. Commun. 9, 536 (2018).
45. Silvério, D. V. et al. Intensification of fire regimes and forest loss in the Território Indígena
do Xingu. Environ. Res. Lett. 17, 045012 (2022).
46. Brienen, R. J. W. et al. Long-term decline of the Amazon carbon sink. Nature 519, 344–348
(2015).
47. Esquivel‐Muelbert, A. et al. Compositional response of Amazon forests to climate change.
Glob. Change Biol. 25, 39–56 (2019).
48. Gatti, L. V. et al. Amazonia as a carbon source linked to deforestation and climate change.
Nature 595, 388–393 (2021).
49. Nepstad, D. et al. Inhibition of Amazon deforestation and fire by parks and Indigenous
lands: inhibition of Amazon deforestation and fire. Conserv. Biol. 20, 65–73 (2006).
50. Botelho, J., Costa, S. C. P., Ribeiro, J. G. & Souza, C. M. Mapping roads in the Brazilian
Amazon with artificial intelligence and Sentinel-2. Remote Sensing 14, 3625 (2022).
51. Matricardi, E. A. T. et al. Long-term forest degradation surpasses deforestation in the
Brazilian Amazon. Science 369, 1378–1382 (2020).
52. Ainsworth, E. A. & Long, S. P. What have we learned from 15 years of free‐air CO2
enrichment (FACE)? A meta‐analytic review of the responses of photosynthesis, canopy
properties and plant production to rising CO2. New Phytol. 165, 351–372 (2005).
53. Kooperman, G. J. et al. Forest response to rising CO2 drives zonally asymmetric rainfall
change over tropical land. Nat. Clim. Change 8, 434–440 (2018).
54. Lapola, D. M., Oyama, M. D. & Nobre, C. A. Exploring the range of climate biome
projections for tropical South America: the role of CO2 fertilization and seasonality: future
biome distribution in South America. Global Biogeochem. Cycles 23, https://doi.org/
10.1029/2008GB003357 (2009).
55. Brienen, R. J. W. et al. Forest carbon sink neutralized by pervasive growth-lifespan
trade-offs. Nat. Commun. 11, 4241 (2020).
56. Lammertsma, E. I. et al. Global CO2 rise leads to reduced maximum stomatal conductance
in Florida vegetation. Proc. Natl Acad. Sci. USA 108, 4035–4040 (2011).
57. Terrer, C. et al. Nitrogen and phosphorus constrain the CO2 fertilization of global plant
biomass. Nat. Clim. Change 9, 684–689 (2019).
58. Ellsworth, D. S. et al. Elevated CO2 does not increase eucalypt forest productivity on a
low-phosphorus soil. Nat. Clim. Change 7, 279–282 (2017).
59. Quesada, C. A. et al. Basin-wide variations in Amazon forest structure and function are
mediated by both soils and climate. Biogeosciences 9, 2203–2246 (2012).
60. Flores, B. M. et al. Soil erosion as a resilience drain in disturbed tropical forests. Plant Soil
https://doi.org/10.1007/s11104-019-04097-8 (2020).
61. Longo, M. et al. Ecosystem heterogeneity and diversity mitigate Amazon forest resilience
to frequent extreme droughts. New Phytol. 219, 914–931 (2018).
62. Levine, N. M. et al. Ecosystem heterogeneity determines the ecological resilience of the
Amazon to climate change. Proc. Natl Acad. Sci. USA 113, 793–797 (2016).
63. Staver, A. C. et al. Thinner bark increases sensitivity of wetter Amazonian tropical forests
to fire. Ecol. Lett. 23, 99–106 (2020).
64. Mattos, C. R. C. et al. Double stress of waterlogging and drought drives forest–savanna
coexistence. Proc. Natl Acad. Sci. USA 120, e2301255120 (2023).
65. Flores, B. M. et al. Floodplains as an Achilles’ heel of Amazonian forest resilience.
Proc. Natl Acad. Sci. USA 114, 4442–4446 (2017).
66. Marengo, J. A. & Espinoza, J. C. Extreme seasonal droughts and floods in Amazonia:
causes, trends and impacts. Int. J. Climatol. 36, 1033–1050 (2016).
67. Boers, N., Marwan, N., Barbosa, H. M. J. & Kurths, J. A deforestation-induced tipping point
for the South American monsoon system. Sci. Rep. 7, 41489 (2017).
68. Alexander, C. et al. Linking Indigenous and scientific knowledge of climate change.
BioScience 61, 477–484 (2011).
69. Ford, J. D. et al. The resilience of Indigenous peoples to environmental change. One Earth
2, 532–543 (2020).
70. Cooper, G. S., Willcock, S. & Dearing, J. A. Regime shifts occur disproportionately faster in
larger ecosystems. Nat. Commun. 11, 1175 (2020).
71. Drijfhout, S. et al. Catalogue of abrupt shifts in Intergovernmental Panel on Climate
Change climate models. Proc. Natl Acad. Sci. USA 112, E5777–E5786 (2015).
72. Salazar, L. F. & Nobre, C. A. Climate change and thresholds of biome shifts in Amazonia:
climate change and Amazon biome shift. Geophys. Res. Lett. 37, https://doi.org/10.1029/
2010GL043538 (2010).
73. Jones, C., Lowe, J., Liddicoat, S. & Betts, R. Committed terrestrial ecosystem changes due
to climate change. Nat. Geosci. 2, 484–487 (2009).
74. Schellnhuber, H. J., Rahmstorf, S. & Winkelmann, R. Why the right climate target was
agreed in Paris. Nat. Clim. Change 6, 649–653 (2016).
75. Chai, Y. et al. Constraining Amazonian land surface temperature sensitivity to precipitation
and the probability of forest dieback. npj Clim. Atmos. Sci. 4, 6 (2021).
76. Brando, P. M. et al. Abrupt increases in Amazonian tree mortality due to drought-fire
interactions. Proc. Natl Acad. Sci. USA 111, 6347–6352 (2014).
77. Berenguer, E. et al. Tracking the impacts of El Niño drought and fire in human-modified
Amazonian forests. Proc. Natl Acad. Sci. USA 118, e2019377118 (2021).
78. Staal, A. et al. Hysteresis of tropical forests in the 21st century. Nat. Commun. 11, 4978
(2020).
79. Staver, A. C., Archibald, S. & Levin, S. A. The global extent and determinants of savanna
and forest as alternative biome states. Science 334, 230–232 (2011).
80. Malhi, Y. et al. Exploring the likelihood and mechanism of a climate-change-induced
dieback of the Amazon rainforest. Proc. Natl Acad. Sci. USA 106, 20610–20615 (2009).
81. Nobre, C. A. et al. Land-use and climate change risks in the Amazon and the need of a novel
sustainable development paradigm. Proc. Natl Acad. Sci. USA 113, 10759–10768 (2016).
82. Burton, C. et al. South American fires and their impacts on ecosystems increase with
continued emissions. Clim. Resil. Sustain. 1, e8 (2022).
83. Smith, C. C. et al. Old-growth forest loss and secondary forest recovery across Amazonian
countries. Environ. Res. Lett. 16, 085009 (2021).
84. Brando, P. M. et al. Prolonged tropical forest degradation due to compounding disturbances:
Implications for CO2 and H2O fluxes. Glob. Change Biol. 25, 2855–2868 (2019).
85. Mesquita, R. C. G., Ickes, K., Ganade, G. & Williamson, G. B. Alternative successional
pathways in the Amazon Basin: successional pathways in the Amazon. J. Ecol. 89,
528–537 (2001).
86. Jakovac, C. C., Peña-Claros, M., Kuyper, T. W. & Bongers, F. Loss of secondary-forest
resilience by land-use intensification in the Amazon. J. Ecol. 103, 67–77 (2015).
87. Barlow, J. & Peres, C. A. Fire-mediated dieback and compositional cascade in an
Amazonian forest. Phil. Trans. R. Soc. B 363, 1787–1794 (2008).
88. Jakovac, A. C. C., Bentos, T. V., Mesquita, R. C. G. & Williamson, G. B. Age and light effects
on seedling growth in two alternative secondary successions in central Amazonia. Plant
Ecol. Divers. 7, 349–358 (2014).
89. Mazzochini, G. G. & Camargo, J. L. C. Understory plant interactions along a successional
gradient in Central Amazon. Plant Soil https://doi.org/10.1007/s11104-019-04100-2 (2020).
90. Schnitzer, S. A. & Bongers, F. Increasing liana abundance and biomass in tropical forests:
emerging patterns and putative mechanisms: Increasing lianas in tropical forests.
Ecology Letters 14, 397–406 (2011).
91. Tymen, B. et al. Evidence for arrested succession in a liana-infested Amazonian forest.
J Ecol 104, 149–159 (2016).
92. da Silva, S. S. et al. Increasing bamboo dominance in southwestern Amazon forests
following intensification of drought-mediated fires. For. Ecol. Manag. 490, 119139 (2021).
93. Carvalho, A. Lde et al. Bamboo-dominated forests of the southwest Amazon: detection,
spatial extent, life cycle length and flowering waves. PLoS ONE 8, e54852 (2013).
94. Adeney, J. M., Christensen, N. L., Vicentini, A. & Cohn‐Haft, M. White‐sand ecosystems in
Amazonia. Biotropica 48, 7–23 (2016).
95. Flores, B. M. & Holmgren, M. White-sand savannas expand at the core of the Amazon after
forest wildfires. Ecosystems 24, 1624–1637 (2021).
96. Veldman, J. W. & Putz, F. E. Grass-dominated vegetation, not species-diverse natural
savanna, replaces degraded tropical forests on the southern edge of the Amazon Basin.
Biol. Conserv. 144, 1419–1429 (2011).
97. Silvério, D. V. et al. Testing the Amazon savannization hypothesis: fire effects on invasion
of a neotropical forest by native cerrado and exotic pasture grasses. Phil. Trans. R. Soc. B
368, 20120427 (2013).
98. Rull, V. A palynological record of a secondary succession after fire in the Gran Sabana,
Venezuela. J. Quat. Sci. 14, 137–152 (1999).
99. Sakschewski, B. et al. Resilience of Amazon forests emerges from plant trait diversity.
Nat. Clim. Change 6, 1032–1036 (2016).
100. Hall, A., Cox, P., Huntingford, C. & Klein, S. Progressing emergent constraints on future
climate change. Nat. Clim. Change 9, 269–278 (2019).
101. Willcock, S., Cooper, G. S., Addy, J. & Dearing, J. A. Earlier collapse of Anthropocene
ecosystems driven by multiple faster and noisier drivers. Nat. Sustain 6, 1331–1342 (2023).
102. Davidson, E. A. et al. The Amazon basin in transition. Nature 481, 321–328 (2012).
103. Hecht, S. B. From eco-catastrophe to zero deforestation? Interdisciplinarities, politics,
environmentalisms and reduced clearing in Amazonia. Envir. Conserv. 39, 4–19 (2012).
104. Hirota, M., Holmgren, M., Van Nes, E. H. & Scheffer, M. Global resilience of tropical forest
and savanna to critical transitions. Science 334, 232–235 (2011).
105. Hawes, J. E. et al. A large‐scale assessment of plant dispersal mode and seed traits across
human‐modified Amazonian forests. J. Ecol. 108, 1373–1385 (2020).
106. Flores, B. M. & Holmgren, M. Why forest fails to recover after repeated wildfires in
Amazonian floodplains? Experimental evidence on tree recruitment limitation. J. Ecol.
109, 3473–3486 (2021).
107. ter Steege, H. et al. Biased-corrected richness estimates for the Amazonian tree flora.
Sci. Rep. 10, 10130 (2020).
108. Poorter, L. et al. Diversity enhances carbon storage in tropical forests: Carbon storage in
tropical forests. Global Ecol. Biogeogr. 24, 1314–1328 (2015).
109. Walker, B., Kinzig, A. & Langridge, J. Plant attribute diversity, resilience, and ecosystem
function: the nature and significance of dominant and minor species. Ecosystems 2,
95–113 (1999).
110. Elmqvist, T. et al. Response diversity, ecosystem change, and resilience. Front. Ecol.
Environ. 1, 488–494 (2003).
111. Esquivel-Muelbert, A. et al. Seasonal drought limits tree species across the Neotropics.
Ecography 40, 618–629 (2017).
112. Estes, J. A. et al. Trophic downgrading of planet Earth. Science 333, 301–306 (2011).
113. Garnett, S. T. et al. A spatial overview of the global importance of Indigenous lands for
conservation. Nat. Sustain. 1, 369–374 (2018).
114. Morcote-Ríos, G., Aceituno, F. J., Iriarte, J., Robinson, M. & Chaparro-Cárdenas, J. L.
Colonisation and early peopling of the Colombian Amazon during the Late Pleistocene and
the Early Holocene: new evidence from La Serranía La Lindosa. Quat. Int. 578, 5–19 (2021).
115. Levis, C. et al. How people domesticated Amazonian forests. Front. Ecol. Evol. 5, 171 (2018).
116. Clement, C. R. et al. The domestication of Amazonia before European conquest. Proc. R.
Soc. B. 282, 20150813 (2015).
Nature | Vol 626 | 15 February 2024 | 563
Analysis
117. Levis, C. et al. Persistent effects of pre-Columbian plant domestication on Amazonian Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
forest composition. Science 355, 925–931 (2017). published maps and institutional affiliations.
118. Coelho, S. D. et al. Eighty-four per cent of all Amazonian arboreal plant individuals are
useful to humans. PLoS ONE 16, e0257875 (2021).
Open Access This article is licensed under a Creative Commons Attribution
119. de Souza, J. G. et al. Climate change and cultural resilience in late pre-Columbian
4.0 International License, which permits use, sharing, adaptation, distribution
Amazonia. Nat. Ecol. Evol. 3, 1007–1017 (2019).
and reproduction in any medium or format, as long as you give appropriate
120. Furquim, L. P. et al. Facing change through diversity: resilience and diversification of plant
credit to the original author(s) and the source, provide a link to the Creative Commons licence,
management strategies during the Mid to Late Holocene Transition at the Monte Castelo
and indicate if changes were made. The images or other third party material in this article are
shellmound, SW Amazonia. Quaternary 4, 8 (2021).
included in the article’s Creative Commons licence, unless indicated otherwise in a credit line
121. Schmidt, M. V. C. et al. Indigenous knowledge and forest succession management in the
to the material. If material is not included in the article’s Creative Commons licence and your
Brazilian Amazon: contributions to reforestation of degraded areas. Front. For. Glob.
intended use is not permitted by statutory regulation or exceeds the permitted use, you will
Change 4, 605925 (2021).
need to obtain permission directly from the copyright holder. To view a copy of this licence,
122. Tomioka Nilsson, M. S. & Fearnside, P. M. Yanomami mobility and its effects on the forest
visit http://creativecommons.org/licenses/by/4.0/.
landscape. Hum. Ecol. 39, 235–256 (2011).
123. Cámara-Leret, R. & Bascompte, J. Language extinction triggers the loss of unique
medicinal knowledge. Proc. Natl Acad. Sci. USA 118, e2103683118 (2021). © The Author(s) 2024

564 | Nature | Vol 626 | 15 February 2024

Methods
Datasets
Our study site was the area of the Amazon basin, considering large areas
of tropical savanna biome along the northern portion of the Brazilian
Cerrado, the Gran Savana in Venezuela and the Llanos de Moxos in
Bolivia, as well as the Orinoco basin to the north, and eastern parts of the
Andes to the west. The area includes also high Andean landscapes with
puna and paramo ecosystems. We chose this contour to allow better
communication with the MapBiomas Amazonian Project (2022; https://
amazonia.mapbiomas.org). For specific interpretation of our results,
we considered the contour of the current extension of the Amazon
forest biome, which excludes surrounding tropical savanna biomes.
We used the Moderate Resolution Imaging Spectroradiometer
(MODIS) Vegetation Continuous Fields (VCF) data (MOD44B version 6;
https://lpdaac.usgs.gov/products/mod44bv006/) for the year 2001
at 250-m resolution124 to reanalyse tree cover distributions within
the Amazon basin, refining estimates of bistability ranges and criti-
cal thresholds in rainfall conditions from previous studies. Although
MODIS VCF can contain errors within lower tree cover ranges and
should not be used to test for bistability between grasslands and
savannas125, the dataset is relatively robust for assessing bistability
within the tree cover range of forests and savannas126, as also shown
by low uncertainty (standard deviation of tree cover estimates) across
the Amazon (Extended Data Fig. 8).
We used the Climate Hazards Group InfraRed Precipitation with
Station data (CHIRPS; https://www.chc.ucsb.edu/data/chirps)127 to
estimate mean annual rainfall and rainfall seasonality for the present
across the Amazon basin, based on monthly means from 1981 to 2020,
at a 0.05° spatial resolution.
We used the Climatic Research Unit (CRU; https://www.uea.ac.uk/
groups-and-centres/climatic-research-unit)128 to estimate mean annual
temperature for the present across the Amazon basin, based on monthly
means from 1981 to 2020, at a 0.5° spatial resolution.
To mask deforested areas until 2020, we used information from
the MapBiomas Amazonia Project (2022), collection 3, of Amazonian
Annual Land Cover and Land Use Map Series (https://amazonia.map-
biomas.org).
To assess forest fire distribution across the Amazon forest biome and
in relation to road networks, we used burnt area fire data obtained from
the AQUA sensor onboard the MODIS satellite. Only active fires with a
confidence level of 80% or higher were selected. The data are derived
from MODIS MCD14ML (collection 6)129, available in Fire Information
for Resource Management System (FIRMS). The data were adjusted to
a spatial resolution of 1 km.
Potential analysis
Using potential analysis130, an empirical stability landscape was con-
structed based on spatial distributions of tree cover (excluding areas
deforested until 2020; https://amazonia.mapbiomas.org) against mean
annual precipitation, MCWD and DSL. Here we followed the method-
ology of Hirota et al.104. For bins of each of the variables, the probabil-
ity density of tree cover was determined using the MATLAB function
ksdensity. Local maxima of the resulting probability density function
are considered to be stable equilibria, in which local maxima below a
threshold value of 0.005 were ignored. Based on sensitivity tests (see
below), we chose the intermediate values of the sensitivity parameter
for each analysis, which resulted in the critical thresholds most similar
to the ones previously published in the literature.
Sensitivity tests of the potential analysis
We smoothed the densities of tree cover with the MATLAB kernel
smoothing function ksdensity. Following Hirota et al.104, we used a flexi-
ble bandwidth (h) according to Silverman’s rule of thumb131: h = 1.06σn1/5,
where σ is the standard deviation of the tree cover distribution and n is
the number of points. To ignore small bumps in the frequency distribu-
tions, we used a dimensionless sensitivity parameter. This parameter
filters out weak modes in the distributions such that a higher value
implies a stricter criterion to detect a significant mode. In the manu-
script, we used a value of 0.005. For different values of this sensitivity
parameter, we here test the estimated critical thresholds and bistability
ranges (Extended Data Table 2). We inferred stable and unstable states
of tree cover (minima and maxima in the potentials) for moving win-
dows of the climatic variables. For mean annual precipitation, we used
increments of 10 mm yr−1 between 0 and 3500 mm yr−1. For dry season
length, we used increments of 0.1 months between 0 and 12 months. For
MCWD, we used increments of 10 mm between −800 mm and 0 mm.
Transition potential
We quantified a relative ecosystem transition potential across the
Amazon forest biome (excluding accumulated deforestation; https://
amazonia.mapbiomas.org) to produce a simple spatial measure that
can be useful for governance. For this, we combined information
per pixel, at 5 km resolution, about different disturbances related to
climatic and human disturbances, as well as high-governance areas
within protected areas and Indigenous territories. We used values of
significant slopes of the dry season ( July–October) mean tempera-
ture between 1981 and 2020 (P < 0.1), estimated using simple linear
regressions (at 0.5° resolution from CRU) (Fig. 1a). Ecosystem stabil-
ity classes (stable forest, bistable and stable savanna as in Extended
Data Fig. 1) were estimated using simple linear regression slopes of
annual rainfall between 1981 and 2020 (P < 0.1) (at 0.05° resolution
from CHIRPS), which we extrapolated to 2050 (Fig. 1b and Extended
Data Fig. 3). Distribution of areas affected by repeated extreme drought
events (Fig. 1c) were defined when the time series (2001–2018) of the
MCWD reached two standard deviation anomalies from historical
mean. Extreme droughts were obtained from Lapola et al.39, based on
Climatic Research Unit gridded Time Series (CRU TS 4.0) datasets for
precipitation and evapotranspiration. The network of roads (paved and
unpaved) across the Amazon forest biome (Fig. 1d) was obtained from
the Amazon Network of Georeferenced Socio-Environmental Informa-
tion (RAISG; https://geo2.socioambiental.org/raisg). Protected areas
(PAs) and Indigenous territories (Fig. 1e) were also obtained from RAISG,
and include both sustainable-use and restricted-use protected areas
managed by national or sub-national governments, together with offi-
cially recognized and proposed Indigenous territories. We combined
these different disturbance layers by adding a value for each layer in the
following way: (1) slopes of dry season temperature change (as in Fig. 1a,
multiplied by 10, thus between −0.1 and +0.6); (2) ecosystem stability
classes estimated for year 2050 (as in Fig. 1b), with 0 for stable forest,
+1 for bistable and +2 for stable savanna; (3) accumulated impacts from
repeated extreme drought events (from 0 to 5 events), with +0.2 for
each event; (4) road-related human impacts, with +1 for pixels within
10 km from a road; and (5) protected areas and Indigenous territories
as areas with lower exposure to human (land use) disturbances, such
as deforestation and forest fires, with −1 for pixels inside these areas.
The sum of these layers revealed relative spatial variation in ecosystem
transition potential by 2050 across the Amazon (Fig. 1f), ranging from
−1 (low potential) to 4 (very high potential).
Atmospheric moisture tracking
To determine the atmospheric moisture flows between the Amazonian
countries, we use the Lagrangian atmospheric moisture tracking model
UTrack132. The model tracks the atmospheric trajectories of parcels
of moisture, updates their coordinates at each time step of 0.1 h and
allocates moisture to a target location in case of precipitation. For each
millimetre of evapotranspiration, 100 parcels are released into the
atmosphere. Their trajectories are forced with evaporation, precipi-
tation, and wind speed estimates from the ERA5 reanalysis product at
0.25° horizontal resolution for 25 atmospheric layers133. Here we use the
Analysis
runs from Tuinenburg et al.134, who published monthly climatological
mean (2008–2017) moisture flows between each pair of 0.5° grid cells
on Earth. We aggregated these monthly flows, resulting in mean annual
moisture flows between all Amazonian countries during 2008–2017.
For more details of the model runs, we refer to Tuinenburg and Staal132
and Tuinenburg et al.134.
Reporting summary
Further information on research design is available in the Nature
Portfolio Reporting Summary linked to this article.
Data availability
All data supporting the findings of this study are openly available and
their sources are presented in the Methods.
124. DiMiceli, C. et al. MOD44B MODIS/Terra Vegetation Continuous Fields Yearly L3 Global
250 m SIN Grid V006. https://doi.org/10.5067/MODIS/MOD44B.006 (2015).
125. Sexton, J. O. et al. Global, 30-m resolution continuous fields of tree cover: Landsat-based
rescaling of MODIS vegetation continuous fields with lidar-based estimates of error. Int. J.
Digital Earth 6, 427–448 (2013).
126. Staver, A. C. & Hansen, M. C. Analysis of stable states in global savannas: is the CART
pulling the horse? – a comment. Global Ecol. Biogeogr. 24, 985–987 (2015).
127. Funk, C. et al. The climate hazards infrared precipitation with stations—a new environmental
record for monitoring extremes. Sci Data 2, 150066 (2015).
128. Mitchell, T. D. & Jones, P. D. An improved method of constructing a database of monthly
climate observations and associated high-resolution grids. Int. J. Climatol. 25, 693–712
(2005).
129. Giglio, L., Schroeder, W. & Justice, C. O. The collection 6 MODIS active fire detection
algorithm and fire products. Remote Sens. Environ. 178, 31–41 (2016).
130. Livina, V. N., Kwasniok, F. & Lenton, T. M. Potential analysis reveals changing number of
climate states during the last 60 kyr. Clim. Past 6, 77–82 (2010).
131. Silverman, B. W. Density Estimation for Statistics and Data Analysis (Chapman & Hall/
CRC Taylor & Francis Group, 1998).
132. Tuinenburg, O. A. & Staal, A. Tracking the global flows of atmospheric moisture and
associated uncertainties. Hydrol. Earth Syst. Sci. 24, 2419–2435 (2020).
133. Hersbach, H. et al. The ERA5 global reanalysis. Q. J. R. Meteorol. Soc. 146, 1999–2049
(2020).
134. Tuinenburg, O. A., Theeuwen, J. J. E. & Staal, A. High-resolution global atmospheric
moisture connections from evaporation to precipitation. Earth Syst. Sci. Data 12,
3177–3188 (2020).
135. Oliveira, R. S. et al. Embolism resistance drives the distribution of Amazonian rainforest
tree species along hydro‐topographic gradients. New Phytol. 221, 1457–1465 (2019).
136. Mattos, C. R. C. et al. Rainfall and topographic position determine tree embolism
resistance in Amazônia and Cerrado sites. Environ. Res. Lett. 18, 114009 (2023).
137. NASA JPL. NASA Shuttle Radar Topography Mission Global 1 arc second. https://doi.org/
10.5067/MEaSUREs/SRTM/SRTMGL1.003 (2013).
138. Hess, L. L. et al. Wetlands of the Lowland Amazon Basin: Extent, Vegetative Cover, and
Dual-season Inundated Area as Mapped with JERS-1 Synthetic Aperture Radar. Wetlands
35, 745–756 (2015).
139. Eberhard, D. M., Simons, G. F. & Fennig, C. D. Ethnologue: Languages of the World.
(SIL International, 2021).
Acknowledgements This work was inspired by the Science Panel for the Amazon (SPA) initiative
(https://www.theamazonwewant.org/) that produced the first Amazon Assessment Report
(2021). The authors thank C. Smith for providing deforestation rates data used in Extended
Data Fig. 5b. B.M.F. and M.H. were supported by Instituto Serrapilheira (Serra-1709-18983)
and C.J. (R-2111-40341). A.S. acknowledges funding from the Dutch Research Council (NWO)
under the Talent Program Grant VI.Veni.202.170. R.A.B. and D.M.L. were supported by the
AmazonFACE programme funded by the UK Foreign, Commonwealth and Development Office
(FCDO) and Brazilian Ministry of Science, Technology and Innovation (MCTI). R.A.B. was
additionally supported by the Met Office Climate Science for Service Partnership (CSSP) Brazil
project funded by the UK Department for Science, Innovation and Technology (DSIT), and
D.M.L. was additionally supported by FAPESP (grant no. 2020/08940-6) and CNPq (grant no.
309074/2021-5). C.L. thanks CNPq (proc. 159440/2018-1 and 400369/2021-4) and Brazil
LAB (Princeton University) for postdoctoral fellowships. A.E.-M. is supported by the UKRI
TreeScapes MEMBRA (NE/V021346/1), the Royal Society (RGS\R1\221115), the ERC TreeMort
project (758873) and the CESAB Syntreesys project. R.S.O. received a CNPq productivity
scholarship and funding from NERC-FAPESP 2019/07773-1. S.B.H. is supported by the Geneva
Graduate Institute research funds, and UCLA’s committee on research. J.A.M. is supported by
the National Institute of Science and Technology for Climate Change Phase 2 under CNPq
grant 465501/2014-1; FAPESP grants 2014/50848-9, the National Coordination for Higher
Education and Training (CAPES) grant 88887.136402-00INCT. L.S.B. received FAPESP grant
2013/50531-0. D.N. and N.B. acknowledge funding from the European Union’s Horizon 2020
research and innovation programme under grant agreement no. 820970. N.B. has received
further funding from the Volkswagen foundation, the European Union’s Horizon 2020 research
and innovation programme under the Marie Sklodowska-Curie grant agreement no. 956170,
as well as from the German Federal Ministry of Education and Research under grant no.
01LS2001A.
Author contributions B.M.F. and M.H. conceived the study. B.M.F. reviewed the literature,
with inputs from all authors. B.M.F., M.H., N.N., A.S., C.L., D.N, H.t.S. and C.R.C.M. assembled
datasets. M.H. analysed temperature and rainfall trends. B.M.F. and N.N. produced the maps
in main figures and calculated transition potential. A.S. performed potential analysis and
atmospheric moisture tracking. B.M.F. produced the figures and wrote the manuscript, with
substantial inputs from all authors. B.S. wrote the first version of the ‘Prospects for modelling
Amazon forest dynamics’ section, with inputs from B.M.F and M.H.
Competing interests The authors declare no competing interests.
Additional information
Supplementary information The online version contains supplementary material available at
https://doi.org/10.1038/s41586-023-06970-0.
Correspondence and requests for materials should be addressed to Bernardo M. Flores or
Marina Hirota.
Peer review information Nature thanks Chris Huntingford and the other, anonymous,
reviewer(s) for their contribution to the peer review of this work. Peer review reports are
available.
Reprints and permissions information is available at http://www.nature.com/reprints.
Extended Data Fig. 1 | Alternative stable states in Amazonian tree cover
relative to rainfall conditions. Potential analysis of tree cover distributions
across rainfall gradients in the Amazon basin suggest the existence of critical
thresholds and alternative stable states in the system. For this, we excluded
accumulated deforestation until 2020 and included large areas of tropical
savanna biome in the periphery of the Amazon basin (see Methods). Solid black
lines indicate two stable equilibria. Small grey arrows indicate the direction
towards equilibrium. (a) The overlap between ~ 1,000 and 1,800 mm of annual
rainfall suggests that two alternative stable states may exist (bistability): a high
tree cover state ~ 80 % (forests), and a low tree cover state ~ 20% (savannas).
Tree cover around 50 % is rare, indicating an unstable state. Below 1,000 mm
of annual rainfall, forests are rare, indicating a potential critical threshold for
abrupt forest transition into a low tree cover state79,104 (arrow 1). Between 1,000
and 1,800 mm of annual rainfall, the existence of alternative stable states
implies that forests can shift to a low tree cover stable state in response to
disturbances (arrow 2). Above 1,800 mm of annual rainfall, low tree cover
becomes rare, indicating a potential critical threshold for an abrupt transition
into a high tree cover state. In this stable forest state, forests are expected to
always recover after disturbances (arrow 3), although composition may
change47,85. (b) Currently, the stable savanna state covers 1 % of the Amazon
forest biome, bistable areas cover 13 % of the biome (less than previous analysis
using broader geographical ranges78) and the stable forest state covers 86 % of
the biome. Similar analyses using the maximum cumulative water deficit (c)
and the dry season length (d) also suggest the existence of critical thresholds
and alternative stable states. When combined, these critical thresholds in
rainfall conditions could result in a tipping point of the Amazon forest in
terms of water stress, but other factors may play a role, such as groundwater
availability64. MODIS VCF may contain some level of uncertainty for low tree
cover values, as shown by the standard deviation of tree cover estimates across
the Amazon (Extended Data Fig. 8). However, the dataset is relatively robust for
assessing bistability within the tree cover range between forest and savanna126.
Analysis
Extended Data Fig. 2 | Changes in dry-season temperatures across the
Amazon basin. (a) Dry season temperature averaged from mean annual data
observed between 1981 and 2010. (b) Changes in dry season mean temperature
based on the difference between the projected future (2021−2050) and observed
historical (1981−2010) climatologies. Future climatology was obtained from
the estimated slopes using historical CRU data128 (shown in Fig. 1a). (c, d) Changes
in the distributions of dry season mean and maximum temperatures for the
Amazon basin. (e) Correlation between dry-season mean and maximum
temperatures observed (1981–2010) across the Amazon basin (r = 0.95).
Extended Data Fig. 3 | Changes in annual precipitation and ecosystem for year 2050, based on significant slopes in (a) and critical thresholds in annual
stability across the Amazon forest biome. (a) Slopes of annual rainfall change rainfall conditions estimated in Extended Data Fig. 1. Data obtained from
between 1981 and 2020 estimated using simple regressions (only areas with Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS),
significant slopes, p < 0.1). (b) Changes in ecosystem stability classes projected at 0.05° spatial resolution127.
Analysis
Extended Data Fig. 4 | Changes in ecosystem stability by 2050 across the
Amazon based on annual rainfall projected by CMIP6 models. (a) Changes in
stability classes estimated using an ensemble with the five CMIP6 models that
include vegetation modules (coupled for climate-vegetation feedbacks) for
two emission scenarios (Shared Socio-economic Pathways - SSPs). (b) Changes
in stability classes estimated using an ensemble with all 33 CMIP6 models for
the same emission scenarios. Stability changes may occur between stable
forest (F), stable savanna (S) and bistable (B) classes, based on the bistability
range of 1,000 – 1,800 mm in annual rainfall, estimated from current rainfall
conditions (see Extended Data Fig. 1). Projections are based on climate models
from the 6th Phase of the Coupled Model Intercomparison Project (CMIP6).
SSP2-4.5 is a low-emission scenario of future global warming and SSP5-8.5 is a
high-emission scenario. The five coupled models analysed separately in (a)
were: EC-Earth3-Veg, GFDL-ESM4, MPI-ESM1-2-LR, TaiESM1 and UKESM1-0-LL
(Supplementary Information Table 1).
Extended Data Fig. 5 | Deforestation continues to expand within the Amazon again, as indicated by increasing rates of annual deforestation in (b).
forest system. (a) Map highlighting deforestation and fire activity between In (b), annual deforestation rates for the entire Amazon biome were adapted
2012 and 2021, a period when environmental governance began to weaken with permission from Smith et al.83.
Analysis
Extended Data Fig. 6 | Environmental heterogeneity in the Amazon forest
system. Heterogeneity involves myriad factors, but two in particular, related
to water availability, were shown to contribute to landscape-scale heterogeneity
in forest resilience; topography shapes fine-scale variations of forest drought-
tolerance135,136, and floodplains may reduce forest resilience by increasing
vulnerability to wildfires65. Datasets: topography is shown by the Shuttle
Radar Topography Mission (SRTM; https://earthexplorer.usgs.gov/)137 at 90 m
resolution; floodplains and uplands are separated with the Amazon wetlands
mask 138 at 90 m resolution.
Extended Data Fig. 7 | The Amazon is biologically and culturally diverse. are being undermined by land-use and climatic changes. Datasets: (a) Amazon
(a) Tree species richness and (b) language richness illustrate how biological Tree Diversity Network (ATDN, https://atdn.myspecies.info). (b) World
and cultural diversity varies across the Amazon. Diverse tree communities and Language Mapping System (WLMS) obtained under license from Ethnologue139.
human cultures contribute to increasing forest resilience in various ways that
Analysis

Extended Data Fig. 8 | Uncertainty of the MODIS VCF dataset across the Amazon basin. Map shows standard deviation (SD) of tree cover estimates from
MODIS VCF124. We masked deforested areas until 2020 using the MapBiomas Amazonia Project (2022; https://amazonia.mapbiomas.org).
Extended Data Table 1 | Examples of positive feedbacks that may affect Amazon forest resilience

Global and regional feedbacks are more likely to propel a large-scale tipping point. Adapted from15.
Analysis
Extended Data Table 2 | Uncertainty ranges in the estimates of critical thesholds for transitions between forest and savanna

A range of possible thresholds results from different values of the sensitivity parameter in the potential analysis, for mean annual precipitation (MAP in mm/year), dry season length (DSL in months)
and maximum climatological water deficit (MCWD in mm). F to S means forest to savanna threshold. S to F means savanna to forest threshold.